AI answers are on top of search results now. What that means for you

If you have looked at your search traffic in the last year and found it drifting down while your rankings held, you are not imagining it, and you are not alone.

Something did change. What is much less clear — and where most of the writing on this subject goes wrong — is how much.

What the page looks like now

The mechanical change is simple. An assembled answer sits at the top of the result page, cites a few sources, and answers the question in place.

What a result page looks like now

leave management system for small business

AI answer

Citedthreeor fourpages

Most people stop here

First ordinary result

The first ordinary result is still there. It is just further down than it used to be, below an answer that may have already told the reader what they wanted to know.

The numbers disagree, wildly

Here is where honesty is more useful than a headline. These are the better-known measurements of what that costs a site:

That is a range from "slightly positive" to "catastrophic" on what sounds like the same question. Any article quoting exactly one of these numbers as the answer is either selling something or has not read the others.

They disagree because they are not measuring the same thing. Some compare pages that have an AI answer against pages that do not — but AI answers appear most often on exactly the questions that never generated clicks anyway, so that comparison is biased before it starts. Others follow the same search terms over time, which is fairer but catches every other change Google made in the same window. Some measure the top-ranking page only; others measure all traffic.

The Pew figures are worth a closer look, because they are the ones people quote most and understand least.

How often people clicked a result

No AI summary shown15%

AI summary shown8%

Bars are drawn relative to each other, not against 100%.

Both numbers are small. Even with no AI answer present, six of seven of those searches ended with nobody clicking a result. This was already the norm. What AI answers did was make a long-standing pattern worse — which is a genuine problem, but a different problem from the one described by "AI is killing search traffic."

What actually changed, in practice

Setting the studies aside, this is the pattern we see in client accounts:

Informational queries lost the most. "What is", "how does", "difference between" — questions with a definitional answer are the ones an AI answer handles completely. If your traffic was mostly this, you have been hit hardest.

Commercial and local queries lost the least. Somebody searching for a supplier, a price, or a business near them still needs to see options, compare them, and make contact. There is no way to satisfy that with a paragraph. This is most of what matters for a business selling a service in a specific city.

Impressions held while clicks fell. The most common signature in Search Console. You still rank; fewer people need to click. Which means a rankings report alone will now tell you almost nothing.

Being cited became worth something. Not as much as a click, but appearing as a source in the answer puts your name in front of the reader at the moment they are deciding who knows what they are talking about.

What is worth doing

Nothing here is exotic. Most of it is what good technical SEO always was, aimed slightly differently.

1. Shift effort towards queries that still convert. Commercial and local intent, comparison pages, pricing pages, service pages. Pages that answer "which one should I choose and who do I call" are far more defensible than pages that answer "what is".
2. Answer the question directly, high up. Answer in the first hundred words, then explain. This suits readers, and it also happens to be what gets extracted as a citation.
3. Publish specifics only you have. Real prices, real timelines, real constraints, from your actual work. A summariser can rewrite generic advice; it cannot invent your numbers. This is also the only durable defence against being outranked by competitors publishing the same generic advice you were.
4. Keep the technical foundations clean. Correct headings, valid structured data, fast pages, crawlable HTML. Being parseable is now the entry fee for being cited at all.
5. Fix your measurement before you change your strategy. If you are still reporting rankings and sessions, you cannot see what is happening. Track clicks and impressions per query group, and track enquiries. Enquiries are the number that pays wages.
6. Do not stop publishing. The sites that lost the most ground are the ones that reacted by publishing less.

What is not worth doing

• Buying a "GEO" or "AI optimisation" package that will not tell you concretely what it changes. Most of what works is ordinary technical and content work under a new name, at a new price.
• Rewriting your whole site for machines. Content written for a summariser reads badly to the person you want to hire you.
• Chasing the biggest trend keywords. We looked at this recently while planning our own writing: the head terms for every hot topic sit at a difficulty where you are competing with global publishers, on questions your customers were never going to buy from. The specific, boring, commercial phrases are where a regional business actually wins.
• Panicking about a traffic drop without checking which queries moved. Sometimes the lost traffic was never going to enquire, and the right response is to stop counting it.

How to check your own position in an afternoon

Open Search Console, set a twelve-month window, and compare clicks against impressions. Then split your queries into two groups: questions, and everything with commercial or local intent.

If impressions are flat and clicks are down on the question group only, the AI answer is doing what it does and your commercial pages are intact. That is an annoying but survivable position. If commercial and local queries are also declining, something else is wrong — and it is more likely a technical or competitive problem than an AI one.

Local search is a different, calmer story

For a business serving a specific place, the map results and local pack still behave much as they did. Somebody looking for a supplier in their city gets shown businesses, not a paragraph. If your enquiries come from your city and the surrounding district, the trend line matters far less to you than to a publisher — and the fundamentals still pay. Our local SEO checklist has not needed a single change because of any of this.

The summary

Search sends fewer clicks for the same visibility, and the effect is concentrated on questions rather than on buying decisions. The response is not a new discipline. It is to move effort towards pages that serve buyers, publish things only you can publish, keep the technical foundations clean, and change what you measure so you can see the difference between traffic and customers.

If your traffic has moved and you want to know which of these explains it, we will look at your Search Console data and tell you what we see — including when the honest answer is that the AI answers are not your problem.